Human pancreatic cancer cell line (Panc 04.03) was obtained from American Type Culture Collection (ATCC) (Manassas, VA, United States) and cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 1% penicillin-streptomycin and 10% fetal bovine serum (FBS) in a humidified atmosphere, 5% CO₂ at 37°C. KRAS^G12D was purchased from Abcam (Cambridge, MA, United States). Hit compounds were purchased from WuXi AppTec.

The virtual screening (VS.) method is designed to search large libraries of compounds in silico. The hit rate is usually much higher than traditional high throughput screening (HTS) (Tang et al., 2006; Shekhar, 2008). The crystal structure of KRAS^G12D (PDB ID: 7EWB) with a high resolution of 1.99 Å was downloaded from the Protein Data Bank (PDB) database and pretreated with the molecular operating environment (MOE) (Scholz et al., 2015) by removing water molecules, adding hydrogen atoms, and optimizing the orientation of hydrogen atoms. Based on the analysis of the interaction between the ligand and the amino acids in the KRAS^G12D active site, pharmacophore features were manually added using the pharmacophore editor of MOE (Nevin et al., 2012).

An in-house chemical database contains 35,000 compounds with diverse structures and covers most of the chemical space, which is conducive to finding new skeleton molecules. The two-dimensional (2D) chemical structures of molecules in the database were firstly converted to 3D structures using the energy minimization algorithm of MOE software (MMFF94x force field) (Zheng et al., 2021). Through the pharmacophore search tool of the MOE (Zhang et al., 2020), all the compounds in the database were then screened according to the pharmacophore model, and the best-match hit compounds were evaluated using the root mean square distance (RMSD) values (The lower the RMSD values, the better the match). Finally, the hit compounds with an RMSD value of less than 0.09 Å were filtered by the molecular docking and the pharmacokinetic (PK) properties.

Molecular docking was carried out through MOE software to evaluate the interaction between the hits of virtual screening and the active sites of KRAS^G12D (Scarpino et al., 2018). The crystal structure file of KRAS^G12D (PDB-ID:7EWB) was prepared for the docking studies where: 1) hydrogen atoms were added and water molecules were removed; 2) the partial charges were computed using Amber99 force field. The bound inhibitor TH-Z835 was used as the template to define the docking-active site in the crystal structure of KRAS^G12D. Molecular docking was performed in the active site using the triangle matcher algorithm. Docking scores were calculated through the dG scoring function of MOE software. The lower docking score indicated better binding affinity between KRAS^G12D and ligand (Zheng et al., 2021).

In silico prediction of the pharmacokinetic (PK) properties of the hit compounds was performed by the ADMETlab tool (Mohammed, 2021). In these calculations, using reference 3144, we evaluated the common PK parameters such as the molecular weight (mol_MW ≤ 800), number of hydrogen bond donors (nHD ≤ 5), number of hydrogen bond acceptors (nHA ≤ 10), number of rotatable bonds (nRot ≤ 10), the aqueous solubility (−7 ≤ logS ≤ 0.5), topological polar surface area (TPSA ≤ 140), caco-2 permeability (CP ≥ -5.15), MDCK permeability, human intestinal absorption (HIA), F_20%, F_30%, volume of distribution (L/kg), T_1/2, rat oral acute toxicity, eye corrosion, and eye irritation.

MST is an effective method to evaluate biomolecular interactions and has been used to study the interactions between binding partners of different molecular sizes (Wienken et al., 2010; Seidel et al., 2013). According to the previously reported method to measure the interaction between small molecule inhibitors and KRAS^G12D protein, the MST assay was performed using Monolith NT.115 instrument (NanoTemp Technologies, GmbH, Munich, Germany) (Wienken et al., 2010). The red fluorescent dye was used for KRAS^G12D fluorescent labeling. Small molecule inhibitors were diluted 1:1 and titrated. The binding buffer (50 mM Tris, 230 mM NaCl) was added to the dilution curve of the small molecule inhibitors and the labeled protein. The final sample was centrifuged at 13,000 rpm for 10 min, and then filled into the capillary tube for MST analysis with 20% LED power and 50% MST power at room temperature. The equilibrium dissociation constant (K _d) was calculated using Nano Temper Analysis software.

According to the previously reported method (Yang et al., 2019), the human pancreatic cancer cell line (Panc 04.03) was seeded at the density of 2.4 × 10⁴ on 96-well plates. After incubation at 37°C with 5% CO₂ for 24 h, the cells were incubated with different concentrations of inhibitors under the same conditions. After 72 h, MTT (0.5 mg/ml, 100 μl) was added after the supernatant was discarded. Then keep the 96-well plate at 37°C for 4 h. After that, abandon the medium and inject 200 μl of dimethyl sulfoxide (DMSO) into each well. The spectrophotometric absorbance of the sample at 570 nm was measured using a Synergy 4 Microplate Reader (BioTek Instruments Inc., United States). All of the compounds were tested three times in each of the cells.

Based on previously reported protocols for investigating inhibition in vivo (Zhou et al., 2022), we injected Panc 04.03 cells (200 μl, 1 × 10⁷ cells) in the subdermal space on the right flank of 6-week-old BALB/c nude mice (Changzhou Cavens Laboratory Animal Co., Ltd, China). All experimental protocols were approved by the Animal Ethics Committee of China Pharmaceutical University (Ethic approval number: 2022–11–001). Once tumors grew to 80–100 mm³, mice were randomly divided into three groups (5 mice per group) and intraperitoneally administered with vehicle, hit compound 3 (3 mg/kg) and hit compound 3 (10 mg/kg) every 4 days for a total of five times. Tumor volume and body weight were measured every 4 days. Tumor volume was calculated using the formula (c × c × d)/2 (c, the smallest diameter; d, the largest diameter).

The high-resolution X-ray structure of KRAS^G12D (PDB ID: 7EWB) in complex with the ligand TH-Z835 was downloaded from the PDB database. The generated structure-based pharmacophore model included four features (Figure 2): two aromatic features (F1 and F2: Aro) and two hydrogen-bond donor features (F3 and F4: Don). The two aromatic features of the pharmacophore model could be mapped onto the naphthalene ring of TH-Z835, which formed hydrophobic interactions with hydrophobic amino acids including Phe78, Met72, Val9 and Ala11. In addition, its hydrogen-bond donor features could match the nitrogen atoms of TH-Z835 that showed multiple hydrogen-bond interactions with Gly60, Glu62, Asp12, and His95. The above results reveal that this model may predict the spatial pharmacophore features of KRAS^G12D inhibitors.

Figure 3 showed the virtual screen scheme that has been successfully used to identify inhibitors in other enzyme systems like Poly (ADP-Ribose) Polymerase-1 (PARP-1) (Zhou et al., 2019) and Tubulin (Zheng et al., 2021). MOE software was used for each step of virtual screening and molecular docking. Firstly, 35,000 compounds in the in-house chemical database were preliminarily screened according to the generated pharmacophore model of KRAS^G12D. Then, 69 compounds with an RMSD value of less than 0.09 Å were selected for the docking-base screening. Before molecular docking, it is very necessary to verify the reliability of docking. In this study, we used the ligand (TH-Z835) of the crystal structure KRAS^G12D-TH-Z835 complex (PDB ID: 7 EW B) as the template for verification. The co-crystallized ligand TH-Z835 was re-docked in the active site of KRAS^G12D (Figure 4A). We can observe that the docking conformation of TH-Z835 was well mapped with the actual conformation in the active site, indicating the good reliability of the docking method. Thus, this verified docking method can be used in virtual screening.

Next, 69 selected compounds of pharmacophore-base screening were docked into the KRAS^G12D binding site. TH-Z835 was used as the positive control and has a docking score of −10.04 kcal/mol. Based on docking scores, four top compounds (namely, hit compounds 1–4) below −10.04 kcal/mol and other three hit compounds (namely, hit compounds 5–7) ranking below them were further selected for predicting the three-dimensional interaction modes. Similar to TH-Z835, hit compounds 1–4 exhibited hydrogen-bond interactions with the key amino acids in the active site including His95, Glu62, Gly60, and Asp12 (Figures 4B–E). However, other hit compounds 5–7 could not form a hydrogen-bond interaction with His95 (Figure 4F). Compared with TH-Z835, halogen atoms in the skeleton of hit compounds 1–4 not only could exhibit stronger hydrophobic interactions with Val9, but also could greatly increase molecular liposolubility and metabolic stability. The chemical structures of all the hit compounds 1–7 were shown in Figure 5. In addition, a good pharmacophore mapping with the hit compounds 1–4 on the model was shown in Figure 6. The naphthalene ring of each hit could match the aromatic features of F1 and F2, while its nitrogen atoms were mapped onto the hydrogen-bond donor features of F3 and F4 respectively.

To provide a reliable pharmacokinetic (PK) property of a compound, in silico prediction of the PK properties of these four hit compounds were carried out using the reference inhibitor 3144 as reference (Table 1). In this calculation, we evaluated these common PK parameters including the molecular weight (mol_MW ≤ 800), number of hydrogen bond donors (nHD ≤ 5), number of hydrogen bond acceptors (nHA ≤ 10), number of rotatable bonds (nRot ≤ 10), the aqueous solubility (−7 ≤ logS ≤ 0.5), topological polar surface area (TPSA ≤ 140), caco-2 permeability (CP ≥ −5.15), MDCK permeability, human intestinal absorption (HIA), F_20%, F_30%, volume of distribution (L/kg), T_1/2, rat oral acute toxicity, eye corrosion, and eye irritation. The result indicated that these parameter values of the hit compounds 1–4 are in the appropriate range. However, the reference inhibitor 3144 exceeded the specified range of nRot and caco-2 permeability respectively. Therefore, based on the molecular interaction and PK analysis, the four candidate hits (hit compounds 1–4) were finally chosen for further biological evaluation.

Microscale Thermophoresis (MST) is a general method for quantifying binding affinity by measuring the equilibrium dissociation constant (K _d) (Bartoschik et al., 2018). The MST binding assays showed that hits 1–4 effectively bound to the KRAS^G12D in the sub-nanomolar range (K _d = 0.13–0.98 nM) and displayed a stronger binding affinity for KRAS^G12D compared with the positive control TH-Z835 (Table 2).

In this experiment, the MTT method was used to detect the effects of hit compounds 1–4 s on the proliferation of human pancreatic cancer cells (Panc 04.03). The results showed that hits 1–4 had a dose-dependent effect on the proliferation of Panc 04.03 cells (Figure 7). In addition, their IC₅₀ values were further calculated. Compared with other hits and the positive control TH-Z835, hit 3 with the IC₅₀ value of 43.80 nM exhibited a stronger anti-proliferative activity on Panc 04.03 cells. Therefore, hit compound 3 was selected as a lead compound and further used for anti-tumor experimental evaluation in vivo.

Given the significant inhibitory effect of hit compound 3 on pancreatic cancer cell growth in vitro, the anti-pancreatic cancer activity of hit compound 3 was evaluated. The mice were randomly divided into three groups (5 mice per group): the vehicle control group, hit compound 3 (3 mg/kg) treated group and hit compound 3 (10 mg/kg) treated group. The hit compound 3 significantly inhibited the growth of Panc 04.03 in a concentration-dependent manner after intraperitoneal injection for 20 days. As shown in Figure 8, the growth rate of xenograft tumors in BALB/c nude mice after hit compound 3 administration was significantly reduced compared with the untreated group. In addition, the significant difference in tumor volume between the untreated and treated groups further demonstrated the inhibitory effect of hit compound 3 on xenograft tumor growth in BALB/c nude mice in vivo.